Fast two-layer nonlinear economic predictive control using machine learning for ORC systems with non-Gaussian disturbances

Abstract

This study proposes a machine learning-based generalized correntropy two-layer nonlinear economic model predictive control (ML-GC-TL-NEMPC) algorithm to improve the computational speed for Organic Rankine Cycle (ORC) systems with non-Gaussian disturbances. In the upper layer, the ratio of the net output power to the total heat transfer area is used as the economic performance index to optimize the optimal reference trajectory. In the lower layer, the generalized correntropy-based MPC (GC-MPC) is used to track the optimal trajectory. To quickly obtain an optimal solution, a Deep Neural Network (DNN) neural network controller is combined with kernel support vector machine (SVM). It learns the possible optimal solution via off-line training. Then it can be quickly implemented on on-line applications, so it can save the optimal calculation time. The proposed ML-GC-TL-NEMPC is applied to ORC. Its computation and performance are faster and better than the traditional two-layer EMPC and single-layer EMPC.